Alumina-supported CO Hydrogenation Catalysts Prepared from Molecular Osmium and Ruthenium Clusters BY HELMUT KNOZINGER AND YAPING ZHAO * Institut fur Physikalische Chemie, Universitat Munchen, Sophienstrasse 1 1, 8000 Munchen 2, Federal Republic of Germany AND BERND TESCHE Fritz-Haber-Institut der Max-Planck-Gesellschaft, Faradayweg 4-6, 1000 Berlin 33, Federal Republic of Germany AND ROGER BARTH, RONALD EPSTEIN,~ BRUCE C . GATES AND JOSEPH P. SCOTT Center for Catalytic Science and Technology, Department of Chemical Engineering, University of Delaware, Newark, Delaware 1971 1, U.S.A. Received 4th June, 1981 0 s catalysts on y-A1203 supports have been prepared from 0 s complexes of varying nuclearity, namely HzOsCj6, OS~(CO)~~, H40~4(C0)12 and OS~(CO)~~. Characterization by infrared and X-ray photoelectron spectroscopy and transmission electron microscopy provides evidence of the stabili- zation of well-defined ensembles of 0 s atoms on the support surface, the ensemble size being deter- mined by the nuclearity of the cluster precursor.The catalysts prepared from HZOsCl6 have a dis- persion comparable with that obtained with OS~(CO)~~, although the presence of smaller ensembles and single atoms cannot be excluded. The ensemble size may influence activity and selectivity for Cz and higher hydrocarbons in CO hydrogenation at atmospheric pressure and temperatures between 530 and 610 K, but the results suggest that the heterogeneity of the 0 s species and the chlorine content of the support also influence the catalyst performance.Data obtained with a more highly dis. persed RujA1203 catalyst prepared from RU~(CO)~~ provide the first quantitative comparison between 0 s and another Group VIII metal catalyst for CO hydrogenation. The 0 s was approximately one order of magnitude less active than the Ru catalyst, but it was more selective for formation of C2 and C3 hydrocarbons. Selective conversion of CO and H2 is a goal of much of the current research in catalysis, since CO and H2 may replace petrochemicals as the basic building blocks of the organic chemicals’ industry. Metals are the principal class of CO hydrogenation catalysts, and selectivities vary widely from metal to metal ; Ni, for example, catalyses formation of methane, and Co catalyses a Schulz-Flory distribution of chiefly straight- chain hydrocarbons, including those with many carbon atoms.The product distributions in metal-catalysed CO hydrogenation are strongly in- fluenced by the location of the metal in the periodic table; metals like Fe and Co, which dissociate CO, are active for hydrocarbon formation, whereas metals like Pd and Pt, which do not dissociate CO, are active for methanol formation.’ It has been suggested that CO hydrogenation reactions involving C-C bond * Permanent address : Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Dalian, t Present address: Stauffer Chemical Co., Dobbs Ferry, New York, U.S.A. Peoples Republic of China.54 co HYDROGENATION CATALYSTS FROM OS, RU CLUSTERS formation require catalysts with neighbouring metal centres, i.e.surfaces or metal clusters., The available data are consistent with this hypothesis, but critical evaluation is required. It has further been suggested that the number of metal centres on a sur- face or cluster influences the chain lengths in the hydrocarbon products, and the data obtained with Ru aggregates encaged in zeolites showed a cut-off in chain length near C2, c6 and Cll, depending on the size of the zeolite cage, which determined the size of the Ru aggregate., The presumed requirement of neighbouring metal centres has motivated the application of metal clusters as catalysts for CO hydrogenation, and there is evidence suggesting that Rh clusters in solution are active for formation of ethylene glycol and other product^.^ The soluble metal clusters, like other molecular catalysts in solution, are attractive in prospect because the discreteness of their structures may imply that they will be selective catalysts.This prospect remains largely untested, primarily because of the difficulties of maintaining cluster structures intact during catalysis. One method of stabilization of metal clusters is to bind them to supports, where they may be held apart from each other in a coordinatively unsaturated (and catalytically active) state. A number of supported metal clusters have been tested as catalystss and a triosmium cluster supported on MgO has been suggested to be active for CO hydrogenation to give alkanes.6 Several examples of highly dispersed oxide-supported metals derived from metal clusters have also been reported to be active catalysts for CO hydrogenation, and some of these have unusual selectivities.For example, Rh prepared from carbonyl clusters on La,O,, TiO, and ZrO, is selective for formation of ethanol,' and Fe on A1203 and MgO is selective for formation of olefins.8 Supported-metal catalysts prepared from triosmium carbonyl clusters have been chosen for further study in this work because they may be " ideal " catalysts, having well-defined structures. The triosmium can be bonded directly to metal-oxide supports, giving supported clusters with unique structure^,^'^^ and these decompose into mononuclear surface carbonyl c~mplexes.~*~-'~ The Al,O,-supported catalysts have been shown to be active for CO hydrogenation, giving methane.ll One of the goals of this research was to investigate the activity and selectivity of CO hydrogenation catalysts prepared from triosmium clusters.To provide a basis for relating the catalytic character to the structure of the metal species, other 0 s cata- lysts were also investigated, including those prepared from a mononuclear complex (H,OsCl,) and from tetranuclear and hexanuclear 0 s clusters. Ru clusters are also of interest for the preparation of supported CO hydrogenation catalysts ; like 0 s clusters, these are evidently broken up into mononuclear specie^'^*'^ but the bonding of the Ru to a SiO, or Al,03 support (presumably through Ru-0 bonds) is not as strong as the bonding of 0 s to these supports, and in the presence of H2 the Ru (in contrast to 0s) undergoes aggregation into ~rystallites.~~ In summary, the research plan was to prepare y-Al,O,-supported 0 s catalysts from OS~(CO)~,, H,OsCl,, H40~4(C0)12 and OS~(CO)~~, and supported Ru catalysts from RU,(CO)~~.These were to be characterized structurally (by infrared and X-ray photoelectron spectroscopy and transmission electron microscopy) and as catalysts for CO hydrogenation. The principal goals were to prepare structurally well-defined supported metals and to determine relations between structure and catalytic activity and selectivity.H . KNOZINGER et al. 55 EXPERIMENTAL MATERIALS AND CATALYST PREPARATION OS~(CO)~~ and R U ~ ( C O ) ~ ~ were obtained from Strem and used without further purification. OS~(CO)~~ was prepared by pyrolysis of Os3(CO)12 at 520 K for 100 h in an evacuated glass tube according to the method described by Eady et al.15 The resulting mixture of 0 s clusters was dissolved in boiling ethyl acetate, and the solution was cooled to room tempera- ture to crystallize unreacted OS~(CO)~~.The supernatant solution contained Os3(CO)12, &5(co)13, OS~(CO)~~, Os7(CO)21 and OS~(CO)~~, which were separated by thin-layer chroma- tography using Kieselgel (Macherey-Nagel, art. no. 80901 3) as the stationary phase and a 1 : 99 mixture of ethyl acetate and cyclohexane as the eluent. The OS~(CO)~~ yield was 3.6%. H40~4(C0)12 was prepared by the method of Lewis et aZ.16 H20sC16-6H20 was ob- tained from Colonial Metals, Elkton, MD. TABLE 1 .---CATALYST ANALYSES a estimated metal content catalyst precursor/support metal content determined by (wt %) elemental analysis (wt %) ~~~ ~~ ~ ~ ~ ~~ * Analyses were performed by Analytische Laboratorien, Engelskirchen, F.R.G.Assuming complete uptake of catalyst precursor from the solution. The relative error is <lo%. The A1203 used as support was A1203,P110C1 from Degussa, F.R.G. This material was prepared by flame hydrolysis of AlC13 and consists of small non-porous spherical particles (5-30 nm in diameter). Its B.E.T. surface area was 100 m2 g-l. In experiments with H40~4(C0)12 a different sample of y-A1203 was used. It was obtained from Ketjen (grade D) and had a surface area of ca. 250 m2 g- l. The cluster-derived catalysts were pre- pared by refluxing the parent carbonyl cluster with the thermally pretreated support (02 at 673 K, followed by evacuation at 110 N m-2 at 673 K) in dry octane (b.p. 398 K) under dry N2, as described previously.11*12 A similar procedure was used with R u ~ ( C O ) ~ ~ and A catalyst was prepared from aqueous H20sC16 and AI2O3 by ion exchange, followed by All the catalysts were analysed for the metal. The results are summarized in table 1. y-Al203. drying and reduction in H2 for 16 h at 423 K then 3 h at 523 K. CHARACTERIZATION METHODS I.R. SPECTROSCOPY Thin self-supporting wafers were pressed and placed in a previously described trans- mission infrared cell, which permits in sit^ heat treatment and chemisorption experiments. The i.r. spectra were recorded on a Perkin-Elmer 225 spectrophotometer. To reduce heat effects induced by the infrared beam, the Globar light source was run at only ca.20% of its maximum power. The spectral slit width in the carbonyl stretching region was typically 3 cm-I.56 co HYDROGENATION CATALYSTS FROM OS, RU CLUSTERS X-RAY PHOTOELECTRON SPECTROSCOPY X-ray source (hv = 1486.6 eV). used as internal standards. low, and repetitive scans were accumulated for the reported spectra. X.p. spectra were recorded on a Leybold Heraeus LHS 10 spectrometer using an A1 Ka Binding energies of 103.0 and 73.5 eV for the Si 2p and A1 2 p levels, respectively, were Because of the low metal loadings, the spectral intensities were TRANSMISSION ELECTRON MICROSCOPY In preparation for examination by high-resolution transmission electron microscopy (TEM), the catalyst powder was subjected to grinding and suspended in hexane. A drop of the suspension was then placed on a 4 nm carbon film mounted on a 1000 mesh Cu grid, and the hexane allowed to evaporate.The micrographs were obtained with a Siemens Elmiskop 102 with an instrumental mag- nification of 377 000 times and an acceleration voltage of 125 keV. To improve the contrast the objective aperture was reduced to ca. 30 pm, which led to a restriction of the spatial frequency spectrum in the electron microscopic image and a restriction of the phase contrast to 10.3 nm. Even with these precautions, there are limitations to be recognized regarding the inter- pretation of the micrographs. The limitations are characteristic of dispersed powder samples, which present variable layer thicknesses and, consequently, varying degrees of defocusing, which produce contrasts which are susceptible to misinterpretation as real structures. Further, local charging of the A1203 is variable, depending on the layer thickness and the contact with the carbon film.Taking these limitations into account, we conclude that the microscopy is capable of detecting 0 s aggregates smaller than 1 nm, but no structural information can be derived for such small species. CATALYTIC REACTION RATE MEASUREMENTS Catalytic reaction experiments were carried out with a steady-state differential flow reactor interfaced to a gas chromatograph. The flow system allowed metering of CO, H2, and He (Matheson UHP grade, further purified by flow through traps to remove traces of oxygen, water and metal carbonyls) to a thermostatted packed-bed reactor which was copper- lined to prevent formation of metal carbonyls.The system was operated at pressures be- tween atmospheric and 3.2 x lo6 N m-2. Pressurereduction downstream of the reactor was achieved by a back-pressure regulator. Details of the system are to be presented elsewhere. l8 Ca. 1 g of catalyst powder (typically) was dispersed between layers of glass wool and loaded into the tubular reactor, which was then packed with glass wool plugs at inlet and exit. The reactor was placed in the flow system and thoroughly purged with helium at lo5 N m-z. The system was then brought to pressure with a continuous helium purge. The reactor was then heated to temperature in a ca. 15 min period, after which CO and Hz reactant gas flows were started. After ca. 1 h steady state was attained and data collection begun.During a catalysis experiment, the reactor temperature, pressure and feed-gas flow rates were held constant, and the product stream (maintained in the vapour phase in a heated exit line) was intercepted periodically with a gas-sampling valve on the gas chromatograph (an Antek 300), which was equipped with a flame-ionization detector and a 3.2 mm x 3 m stainless-steel column packed with 60/80 mesh alumina. The helium carrier-gas flow rate was 18 cm3 min-'. Temperature programming of the column involved an initial 20 min hold at 410 K, heating at 20 K min- ' to 490 K, then a 5 min hold at 490 K. The gas chroma- tograph was calibrated with the following known compounds to determine response factors and elution times : methane, ethane, ethylene, prdpane, propylene and n-butane.Products were identified by their elution times.H. KNOZINGER et al. 57 RESULTS AND DISCUSSION The decomposition of numerous metal carbonyl clusters to form mononuclear carbonyl species on oxide supports has been shown to be typically accompanied by reaction with OH groups of the support and oxidation of the meta1.19*20 The ten- dency of a cluster to undergo decomposition is expected to be indicated by the relative strengths of metal-metal and metal-oxygen bonds, and a summary of relevant litera- ture data (for zero-valent 0 s and Ru) is given in table 2. For 0 s and Ru, the metal- metal bond is the weakest in the cluster carbonyl. 0s-0 bonds are extremely strong, and Ru-0 bonds are significantly weaker.We would therefore expect strong osmium-support interactions in oxide-supported 0 s catalysts. We might expect stable mononuclear osmium species, the presence of which might be associated with the presumably monolayer raft-like structures reported by Sinfelt et al.23*24 on silica supports . TABLE 2.-AVERAGE BOND STRENGTHS IN METAL CARBONYL CLUSTERS (hE) AND BOND DISSOCIATION ENERGIES (D'298) OF DIATOMIC SPECIES AT 298 K (kJ m0l-l) compound bond A B 0'298 ref. - 129 190 - 397 - < 594 117 - 171 - - 292 - 481 - 21 21 21 22 21 21 21 22 The relatively weak interactions between Ru and 0, on the other hand, are ex- pected to allow more facile aggregation of Ru on oxide surfaces. Kuznetsov et CLZ.'~ recently reported the formation of Ru microcrystallites from R U ~ ( C O ) ~ ~ on A1203 after reduction in H2.These microcrystallites, however, could be redispersed into smaller entities when the sample was heated in CO to temperatures >470 K. In contrast, supported 0 s catalysts formed from O S ~ ( C O ) ~ ~ are resistant to metal aggregation and mononuclear carbonyls are the predominant surface species obtained on thermal decomposition of O S ~ ( C O ) ~ ~ on A1203.9*11 Knozinger and Zhao12 recently reported infrared spectra and a model for these surface species. The infrared spectra in the carbonyl region showed two sets of band pairs at 2130 and 2037 cm-l and at 2050 and 1970 ern-', which were assigned to an -OS~~(CO)~ and - O S ~ ~ ( C O ) ~ species, re~pectively.~~~~ These two species could be interconverted without the occurrence of detectable metal aggregation at temperatures up to 770 K.A detailed analysis of the infrared spectra allowed an estimate of bond angles in these species, and it was concluded from the geometries of the complexes and the size of the carbonyl ligand that the minimum 0s-0s distance between two -Os(CO), units was ca. 0.59 nm. This value, compared with the 0s-0s distance of 0.2877 nm25 in O S ~ ( C O ) ~ ~ , clearly confirms the earlier c o n c l ~ s i o n ~ ~ ~ ~ that the original cluster breaks up during heat treatment to give mononuclear carbonyl complexes.58 co HYDROGENATION CATALYSTS FROM OS, RU CLUSTERS Additional characterization of these complexes is provided by the results of this work, including reactivity of the osmium species in the presence of H2, O2 and CO in the temperature regime in which catalytic hydrogenation of CO occurs.c AT A L Y s TS PREP ARE D FROM O S ~ ( C O ) ~ ~ / A ~ ~ O ~ PHYSICAL CHARACTERIZATION The strength of the 0s-O bonds leads us to expect that the 0 s carbonyl species formed from the clusters may be immobile even at elevated temperatures and may retain their initial positions [with a minimal Os-Os distance of ca. 0.59 nm between --OS~~(CO)~ units], thus forming ensembles of three 0 s atoms on the surface. This expectation is reinforced by the results of Deeba et aZ.,6 who observed the decomposi- tion of osmium clusters on MgO followed by the formation of clusters again in the presence of CO. The estimated diameter of an ensemble is ca. 0.98 nm, and the average distance between ensembles on the A1203 support, estimated from the total metal loading, is ca.7 nm. The transmission electron micrograph of plate 1 confirms the existence of these ensembles. Scattering centres of extremely uniform size (< 1 nm) are clearly evident, and we attribute them to the three-atom ensembles, the average distance between them being roughly the predicted value of 7 MI. The osmium formed by cluster decomposition has been inferred by Smith et aZ.? who measured the stoichiometry of the surface reaction, to be divalent. As mentioned above, this conclusion is consistent with the positions of reported infrared carbonyl stretching Further support is provided by the X-ray photoelectron spectra of fig. 1. The dotted lines in the figure represent the binding energies at 50.0 and 52.7 eV of the 4h12 and 4h12 core levels of zero-valent Os, respectively, which have been measured for a mechanical mixture of OS,(CO)~~ and A1203.Spectrum 1 represents the trinuclear cluster HOS~(CO)~~OS~<, which is bound to a Si02 surface via an edge- bridging oxygen ligand.g-12 This compound was synthesized according to a pre- viously described method and characterized by its infrared carbonyl spectr~m.~-l~ The spectrum can be deconvoluted into a pair of doublets at 49.9 and 53.7 eV and 51.5 and 54.2 eV, which represent, respectively, the 4h12 and 4f12 binding energies of a single osmium bonded to CO ligands only and the two edge osmium atoms bridged by the oxygen in the HOs3(CO)loOSi< cluster. The structureless band (spectrum 2) in fig.1 was obtained with the freshly prepared O S ~ ( C O ) ~ ~ / A ~ ~ O ~ catalyst (after exposure to air), and spectrum 3 is typical of the -OS~~(CO)~ species formed from it. This latter spectrum shows the unresolved 4f doublet (4f,7/2 binding energy at 52.3 ev). The width of these bands (f.w.h.m. - 7.2-7.9 eV for the doublet) is attributed to the surface heterogeneity. The weak shoulders at low binding energies can pre- sumably be explained as A1 2s(Ka) and 0 s 4f(Ka3,J X-ray satellites. The most important result provided by the X.p. spectra is the shift of the 4f712 levels in the oxidized catalyst toward higher binding energies; the shift is 2.3 eV com- pared with the zero-valent 0 s reference value, which is in good agreement with the oxidation state +2 attributed to the mononuclear 0 s species.After treatment in H2 at 650 K, the binding energies were shifted back toward the reference values for zero- valent 0 s (fig. 1). This result clearly indicates a reduction of the original +2 oxidation state, but the true oxidation state of the reduced osmium cannot be determined from these data, since the sample had to be transferred through the atmosphere to the X.P.S. apparatus from the infrared cell where it underwent reduction. (The same wafer was used for X.P.S. and TEM after characterization by infrared spectroscopy.)PLATE 1 .-Transmission electron micrograph of OS~(CO),~/AI~O~ after formation of -OS"(CO)~ species (some 0 s scattering centres are indicated by arrows). [To face page 58PLATE 2.-Transmission electron micrograph of Os3(CO),,/Al,O3 uscd as a catalyst for CO hydro- genation. Some 0 s scattering centres are indicated by arrows.[To face page 59H , KNOZINGER et al. 59 The question now arises whether metal aggregation had occurred during the high- temperature (650 K) reduction. Only scattering centres of uniform size (<I nm), indistinguishable from those observed in the oxidized samples, can be discerned in the electron micrograph (not shown). No larger metal particles appear to have been formed.* Therefore, we infer that the structural model described above for the oxi- dized samples can equally well be applied to the reduced catalysts. FIG. 1.-X.p. prepared, (3) ! I oso 4f5& 4f7* spectra of supported O S ~ ( C O ) ~ ~ : (1) HOS~(CO)~~OS~<, (2) OS~(CO),~/A~~O~ freshly O S ~ ( C O ) ~ ~ / A ~ ~ O ~ after formation of -OS~~(CO)~, (4) sample of spectrum (3) after reduction in Hz at 650 K.The interaction of the supported 0 s with H2, O2 and CO has been studied by infrared spectroscopy in the carbonyl stretching region. The three characteristic bands at 2128-2130, 2038-2050 and 1965-1970 cm-' [which were observed for the oxidized samples containing -OSI~(CO)~ and -OS~I(CO)~ species] were eliminated on reduction in H2 at 650 K, and a new band pair appeared at 2025 and 1925 cm-l (fig. 2, spectrum l), which is typical of the reduced state of the catalyst. The band pair may be assigned to the symmetric and antisymmetric stretching modes of an -Os(CO), species, the 0 s atom being in a low oxidation state.(Note that no CO was admitted after reduction.) Admission of O2 at room temperature led to a slight shift towards higher wavenumbers, while O2 treatment at 470 K led to reoxidation and re- * In contrast, Smith et 0L9 stated that 0 s particles could be obtained by reduction for >24 h at 470 K, but details of their experiment and the metal loading of their catalyst are lacking.60 co HYDROGENATION CATALYSTS FROM OS, RU CLUSTERS stored the typical band pair at 2045 and 1965 cm-l characteristic of the -OS~~(CO)~ species (fig. 2, spectrum 3). Subsequent exposure to a CO atmosphere at 620 K produced a complete recarbonylation accompanied by the reappearance of the original set of bands at 2128, 2038 and 1965 cm-l (fig. 2, spectrum 4). These results demon- wavenumber/cm - FIG.2.-Infrared spectra of OS~(CO)~Z/AI~O~: (1) after reduction in H2 at 9.3 x lo4 N m-2 and 650 K for 4 h, (2) after subsequent exposure to O2 at 9.3 x lo4 N rn-' and 298 K for 1 h, (3) after heating in O2 at 9.3 x lo4 N m-2 and 473 K, (4) after subsequent exposure to CO at 1.3 x lo4 N m-2 and 625 K for 4 h. strate that the 0 s species can be reduced and reoxidized reversibly without any detect- able metal aggregation ; electron microscopy provided confirmation of this con- clusion. A spectrum representative of -OS"(CO)~ mixed with -OS~~(CO), is shown in fig. 3 (spectrum 1). Spectrum 2 is that of the sample reduced in H,. Exposure of this reduced sample to CO at 1.3 x lo4 N m-2 and room temperature produced band shifts toward higher wavenumbers and a shoulder at 2040 cm-l. Heating the sampleH .KNOZINGER et al. 61 6 I 0 2, I I I I I I I I I 1 4 /' I/ 00 * w 0 0 wc.l wavenumber/cm-' FIG. 3.-Infrared spectra of OS~(CO)~~/AI~O~: (1) sample treated in CO at 9.3 x lo4 N m-2 and 723 K for 15 h, (2) after reduction in H2 at 9.3 X lo4 N m-' and 650 K for 4 h, (3) after exposure of(2) to CO at 1.3 x lo4 N m-' and 298 K for 15 h, (4) after treatment in CO at 1.3 x lo4 N m-2 and 625 K for 4 h. in CO at 1.3 x lo4 N m-2 and 620 K essentially restored the original spectrum characteristic of the -OS~~(CO)~ species. We therefore conclude that, even at room temperature, a partial reoxidation of the 0 s species occurs and that CO treatment at elevated temperatures would lead to a complete reoxidation.Analogous observations have been reported by Primet26 for Rh in zeolites. To account for this observation, one must assume either a simple dissociative chemisorption of CO or CO disproportionation with subsequent dissociation of CO,: CO(g) - C(a> + O(a) 2CO(g) - C(a) + co2 CO&) - CO(g) + O(a>.62 co HYDROGENATION CATALYSTS FROM OS, RU CLUSTERS 0 s has been classified as a metal that would adsorb CO only ass~ciatively.~~*~* It is clear that this classification does not pertain to our supported 0 s under the condi- tions investigated. Since the formation of a surface carbide via CO dissociation and/ or disproportionation is considered to be a necessary step for catalytic methanation and Fischer-Tropsch ~ y n t h e s i s , ~ * ~ * ~ ~ - ~ ~ we infer that the supported 0 s samples may be appropriate model catalysts for CO hydrogenation, as is discussed further below.The structural model presented above fails to account for dissociative chemisorp- tion of CO on the reduced catalysts, since an 0s-0s distance of 0.59 nm is too great to allow CO dissociation through the following structure: c-0 / \ 0 s 0s. To account for the CO dissociation, we suggest that the 0 s atoms were closer together than 0.59 nm, and we infer that it is important that no spectroscopic evidence was obtained for the formation of tricarbonyl species on the reduced samples. The steric repulsion between carbonyl ligands in neighbouring -OS~~(CO)~ complexes, which is considered to be chiefly responsible for the large separation between them, is expected to be much weaker in the reduced species.We suggest that the 0s-0s distance between neighbouring dicarbonyl species may approach the 0s-0s distance in the OS~(CO)~~ cluster. This is possible if neighbouring 0 s species are placed into adjacent sites bridging two or three surface oxygen ions on (100) and (1 11) faces of the A1203 support (note that the size of the 0 s atom is very nearly the same as that of an oxygen ion), while only every second site can be occupied by the oxidized species.12 If one allows for relatively small local reorganizations within the ensembles, depending on the oxidation state and degree of carbonylation of the 0 s atoms, one can explain how dissociative chemisorption of CO might occur and provide the surface carbide inter- mediate necessary for methanation and Fischer-Tropsch reactions.CATALYTIC HYDROGENATION OF co The catalysts prepared from O S ~ ( C O ) ~ ~ have been found to be active for hydro- genation of CO to give alkenes and alkanes. To simplify the product analysis, reaction conditions were chosen so that only low conversions were observed (< 1 %, assumed to be differential), and only C1-C4 hydrocarbons were found among the organic products. Representative conversion data from an experiment carried out at a pressure of 7.9 x lo5 N me2 are shown in fig. 4. These results indicate a slow loss of catalytic activity (presumably resulting from the formation of carbonaceous deposits on the surface) and a nearly time-invariant product distribution. Similar and more thorough results were obtained at a pressure of 1.0 x lo5 N m-2.The rate data and the product distribution data (giving selectivities for C , and C3 products) are summarized in table 3. These data (consistent with those of fig. 4) show that the supported 0 s catalyst was active for formation of CH4, with lower rates of formation of C2H4, C2H6, C3H6 and C3H8. At the lower pressure, the catalyst deactivation was so slow that reaction rates characteristic of the fresh catalyst could be determined at more than one temperature. Representative data are shown in the Arrhenius plot of fig. 5, where they are compared with data obtained with the catalyst prepared from H20sCl,. The Arrhenius parameters are included in table 3. Infrared spectra of the used catalyst were virtually indistinguishable from those of the fresh catalyst in the oxidized form, confirming an earlier. qualitative rep0rt.l'H.KNOZINGER et al. 63 These results suggest that the ensembles of 0 s on the A1203 were catalytically active for formation of methane and Fischer-Tropsch products. The electron micrograph of the used catalyst (plate 2) is consistent with this interpretation; there is no evidence of larger 0 s aggregates than those present in the fresh catalyst. The only change evident in the micrograph is consistent with the presumed presence of carbonaceous deposits on the A1203 support. - 0 - - _ c3 10-51 I I I I ! I I I 0 10 20 30 40 50' time on stream/h FIG. 4.-CO conversion catalysed by O S ~ ( C O ) ~ ~ / A ~ ~ O ~ : reactant, He:H2:C0 = 1 : 3: 1, total pres- sure: 7.9 x lo5 N m-2, temperature: 573 K, mass of catalyst: 0.1651 g, 0 s content: 0.24 wt %, In summary, all the results point to the ensembles of three.Os atoms as the catalysts for CO hydrogenation.There is no evidence of the existence of larger 0 s aggregates. These results therefore appear to be important in providing the first evidence of CO hydrogenation in the presence of a structurally well-defined supported-metal catalyst. The results suggest that structure-property relations might be obtained by preparation of CO hydrogenation catalysts from 0 s complexes of various nuclearities to produce enembles of various sizes. Such catalysts are reported in the following paragraphs. feed ffow rate: 166 scc min-' (T = total conversion). CATALYSTS PREPARED FROM H40~4(C0)12/A1203 The samples prepared from the tetraosmium cluster have been characterized in a preliminary way with infrared spectroscopy.The A1203-supported 0 s (handled in the absence of air) presented a spectrum with bands at 2123 (vw), 2091 (w), 2058 (s), 2027 (s) and 2012 (s) cm-'; this spectrum is clearly different from that of the sample prepared from O S ~ ( C O ) ~ ~ and A1203, different from that of H40~4(C0)12 itself [2119 (vw), 2065 (vs), 2021 (s), 198O(w)] and different from that of CO adsorbed onTABLE 3. - CATALYTIC ACTIVITIES AND SELECTIVITIES IN CO HYDROGENATION feed composition activity a selectivities apparent activation energies,c Eact molar ratio catalyst lo4 rcH4 rC2H4 rC2H4 + rC2H6 rC3H6 + rC3H8 rC4H10 (CH4) (C2H4) (C2H6) (C3H8) (C4H10) He: H2 : CO YCH4 YCH4 TCHq YCH4 1:3:1 os3(co)i2/ 1:3:1 H20sCl,/ A1203 1.5 0.17 0.20 0.04 - w100 115 - - - A1203 25 0.03 0.05 0.003 - 139 130 - 110 - M 120 A1203 46 0.06 0.08 0.008 0.001 138 - 180 - 7.2 0.05 0.06 - - 161 140 - - - 1:3:1 RU3(C0)12/ 3:l:l H2OSCl6/ A1203 3:l:l RU3(C0)12, - - - - - 120 A1203 8.7 0.07 0.08 - ~~ ,I Rate r of CH, formation, molecules (metal atom s)-l.Data were extrapolated slightly to 606 K. Reaction at 1.0 x lo5 N m-2. ' Rates in molecules of hydrocarbon product (metal atom s)-'. Apparent activation energies in kJ mol-l, from plots of log r against inverse temperature.H . KNOZINGER et al. 65 0 s These results suggest that the osmium was bonded to the support in a highly dispersed form, possibly as tetraosmium clusters. The molecular cluster H40s4(CO)12 is stable in air at room temperature, but the A1203-supported species underwent structural changes in air, the aforementioned car- bony1 bands disappearing and a complex new set of bands appearing.1.7 1.8 1.9 1 0 3 ~ 1 ~ FIG. 5.-Arrhenius plot: initial rates of formation of CH4 from CO + H2 at 1.0 X lo5 N m-2, providing a comparison of the two supported 0 s catalysts. 0, 1:3:1 He:H2:CO; 0, 3 : l : l He: H2: CO; (-) H20sC16/A1203; (- - -) O S ~ ( C O ) ~ ~ / A ~ ~ O ~ . When the initially supported 0 s was heated in vacuum from room temperature to 508 K, the bands disappeared, and a new set of bands formed at ca. 21 15 (w), 2030 (m) and 1960 (w) cm- ; this new set of bands is indicative of a mononuclear 0 s carbonyl and is interpreted as evidence that the initial cluster species broke up to form en- sembles of mononuclear 0 s complexes, perhaps four-atom ensembles.The samples prepared from H40~4(C0)12 were tested as catalysts at 473 K and 3.2 x lo6 N m-2. The catalyst deactivation was negligible for 140 h of operation in the flow reactor. The product distribution was nearly independent of the H2:C0 ratio in the feed (table 4), and the selectivity for C2 formation was large in comparison with that of the sample prepared from O S ~ ( C O ) ~ ~ (table 3), which, however, was investigated under different conditions. The difference in the selectivities of the catalysts prepared from tri- and tetra- nuclear 0 s clusters suggests that the size of the ensemble may influence the product distribution, but this is no more than a speculation because the two catalysts were tested under different conditions.66 co HYDROGENATION CATALYSTS FROM OS, RU CLUSTERS TABLE 4.<ATALYTIC ACTIVITY OF H~OS~(CO)I~/A~~O~ AT 473 K AND 3.2 X 106 N m-2 H2: CO molar ratio in feed product 3:l 2: 1 1:l CHI 7.9 x lo-6 7.8 x 7.6 x C2H6 + C2H4 4.6 x 4.6 x 4.7 x C3H8 + C3H6 - 1.7 x 1 0 - 7 1.6 x 1 0 - 7 a Reaction rates: molecules of hydrocarbon formed per 0 s atom per second.Conversions of CO varied from 4 x to 9 x lo-'. The catalyst was loaded into the reactor without contacting air, and was brought to the reaction temperature in flowing CO. CAT A LY ST s PREPARED FROM OS~(CO)~~/AI~O~ The thermal decomposition of O S ~ ( C O ) ~ ~ on A1,03 supports has been described by Smith et aZ.,9 whose carbonyl infrared spectra were virtually identical to those charac- teristic of the mononuclear decomposition products of the supported triosmium cluster.We have confirmed this result and we infer that the same mononuclear -OS~~(CO)~ and --OS~~(CO), species mentioned previously were formed from the hexanuclear cluster. However, the ensemble should now be larger, consisting of six atoms. A transmission electron micrograph of the OS~(CO),~/AI~O~ after thermal decomposition (not shown) shows scattering centres of uniform size of ca. 1.2 nm. These results strongly support the proposed model of ensemble formation and demon- strate that unique catalysts containing 0 s ensembles of well-defined size can be tailored using osmium carbonyl clusters of varying nuclearity as the catalyst precursors.CATALYSTS PREPARED FROM H2OsC&/Al,O, A sample prepared by bringing aqueous H2OSCI6 in contact with y-A1,03 had a carbonyl infrared spectrum closely resembling that of the mononuclear species ob- tained from the carbonyl clusters, as observed previous1y.l' A similar SO,-supported sample was studied by Prestridge et aZ.24 by electron microscopy. Some of the 0 s particles appeared to be present in " rafts ", presumably monolayers with an average width of ca. 1.2 nm. EXAFS showed that the average 0s-0s distance was the same as that in metal films or clusters (0.78 nm).33 The 0 s in these samples was presumably zero-valent and had no adsorbed carbonyl ligands. On Al,03 supports the formation of highly dispersed structures might be expected, and these might have a less dense packing on Al,03 than on Si02 because of a stronger 0s-0 interaction.Plate 3 shows an electron micrograph of the catalyst prepared from H,OsCl,. Highly dispersed ensembles are evident, being no larger than the three-atom ensembles shown in plate 1. We recognize that there may also be isolated mononuclear com- plexes too small to be observed by the microscope. This catalyst was also active for CO hydrogenation, giving alkanes and alkenes. The rate data are summarized in table 3, with some data also shown in the Arrhenius plot of fig. 5. The activity of this catalyst for hydrocarbon formation is more than one order of magnitude greater than that of the catalyst prepared from triosmium clusters, whereas the selectivity for C2 and C3 formation is an order of magnitude less.The used catalyst has been characterized by infrared spectroscopy (fig. 6). A comparison with the spectrum of the fresh catalyst (which was nearly indistinguishablePLATE 3.-Transmission electron micrograph of H20sC1,/A1203 after reduction under conditions described in the Experimental section. [To face page 66H. KNOZINGER et al. 67 d ._ M -g C c! Y wavenumber/cm - FIG. 6.-Infrared spectra of catalyst H20sC16/A1203 after use in CO hydrogenation under conditions given in table 3: (I) used catalyst exposed to CO at 8 x lo5 N m-2 and 298 K, (2) after CO treatment at 8 x lo5 N m-’ and 473 K. from spectrum 4 in fig. 3) provides no indication of structural changes during the catalysis. We infer that this catalyst, like the one prepared from OS~(CO)~~, was stable and, in particular,.that no aggregation of the 0 s took place.The reasons for the differences in performance of the two 0 s catalysts are not clear. We recall that the catalyst prepared from H20sC16 contained 3.5 times as much 0 s as the catalyst prepared from Os,(CO),,; the former catalyst also incorporated C1- (derived from H20sC16), which is expected to have increased the acidity (including the Lewis of the support. The analytical values for C1 were 1.23% and 0.25%, respectively, for the two catalysts. A role of the support cannot be ruled out: for example, the support may alter the specific properties of the osmium species, or Lewis acid centres (coordinatively unsaturated A13+ ions) in the support surface could play a role in the catalysis, such as aiding in the dissociation of CO.Since the catalyst prepared from H,0sC16 is expected to be heterogeneous, in- corporating mononuclear surface complexes as well as ensembles, one might speculate68 co HYDROGENATION CATALYSTS FROM OS, Ru CLUSTERS that the mononuclear 0 s complexes are involved in CO dissociation, perhaps with the assistance of AP+ sites, as mentioned above. c A T A L Y s T s PREP A RED FR o M R U ~ ( C O ) ~ ~ / A ~ ~ O ~ and since the thermal decomposition of Ru carbonyl clusters on A1203 has been investigated by ~ t h e r ~ , ~ ~ ~ ~ ~ ~ ~ ~ we discuss the infrared spectra of the RU~(CO)~~/A~,O~ sample only briefly. Fig. 7 shows a typical spectrum of the decomposition product (spectrum 1).Bands can be discerned at 2160, 2140, 2072 and 2002 cm-l. In addition, an asymmetry is evident at ca. 2050 cm-l on the low-wavenumber flank of the Since infrared spectra of CO chemisorbed on supported Ru have been wavenumber/cm-' FIG. 7.-Infrared spectra of RU~(CO)~~/AI~O~ : (1) fresh catalyst (exposed to air) after evacuation (< 1 x lo2 N m-2) for 2 h, (2) after treatment in Hz at 5.3 X lo5 N m-2 and 473 K for 15 h, (3) after treatment in H2 at 5.3 x lo5 N m-z and 573 K for a period of 1 h, (4) for a period of 6 h, (5) for a period of 26 h.H. KNOZINGER et al. 69 band at 2073 cm-', and a shoulder is evident at ca. 1970 cm-'. It can be ~ h o w n ' ~ . ~ ~ that this spectrum must be composed of the contributions from several mononuclear Ru carbonyl species.We assign the band pair at 2140 and 2072 cm-l as a -Ru(CO), species, and two types of -Ru(CO), species are likely to exist with carbonyl stretching bands at ca. 2070 and 2002 cm-l and at 2050 and 1970 cm-l, respectively. By ana- logy with the O S ~ ( C O ) ~ ~ / A ~ ~ O ~ system, we infer that the Ru atoms are in a positive oxidation state in these mononuclear surface species, in agreement with the con- clusions of Smith et aZ.*O This inference is also substantiated by the close similarity of the carbonyl spectra to those of known molecular RulI c~mplexes.'~ The difference between the two proposed dicarbonyl species may be attributed to different oxidation states of Ru and different coordination by surface oxygens. The bands at 2160 and 2140 cm-l vanished, the intensities of the bands at 2072 and 2002 cm-l decreased, the intensity of the band at 1972 cm-l increased and a new band at 2053 cm-l was clearly resolved.This band, together with contributions to the 1970 cm-l band, suggests the formation of a -Ru(CO), species with Ru in a lower oxidation state. However, reduction under more severe conditions (573 K) led to a gradual disappearance of all carbonyl bands with the exception of the band at 1972 cm-l (fig. 7, spectra 3-5). This band could not be eliminated by evacuation or treatment in H2 at 573 K. Only treatment in O2 at 473 K led to complete removal of the band, and subsequent CO adsorption restored the original spectrum of the oxidized sample (fig. 7, spectrum 1). A carbonyl spectrum of this type with a single band at ca.1970 cm-l has not been reported before. The band occurs at a surprisingly low frequency and an unequivocal assignment is not possible. Two explanations are considered plausible : (1) The band is indicative of a terminal CO ligand bonded to a low-valent Ru atom. The low frequency can then be explained only if the Ru atom bears additional electropositive ligands. These may be carbide ligands resulting from CO dissociation, which are known to produce a red-shift of carbonyl stretching bands.36 This explanation implies that somehow CO dissociation occurs on the surface. (2) Alternatively, the band might be assigned to a bridging carbonyl, although it is present at a relatively high frequency for such a species. Both interpretations could be accounted for by ensembles of Ru atoms or Ru particles with metal-like Ru-Ru distances.We might therefore suggest more pronounced aggregation of Ru than of Os, presumably due to the weaker Ru-0 interaction. Kuznetsov et aZ.13 and Ugo et aL3' reported the appearance of Ru micro- crystallites upon reduction of the supported cluster-derived species in H2. The catalytic reaction experiments confirm the activity of Ru for CO hydrogena- tion to give hydrocarbons (table 3). The catalytic activity is represented per Ru atom, even though the dispersion of Ru is not known. We presume that it is nearly unity, since the infrared spectra provide no evidence of CO adsorbed on metal in the used catalysts. Represented in this way, the activity is approximately one order of mag- nitude greater than that of the 0 s catalyst prepared from OS~(CO)'~.These results provide the first quantitative comparison of the CO hydrogenation activities of these two metals." The data of table 3 also show that the Ru catalyst was less selective than the catalyst prepared from OS,(CO),~ for formation of C , and C3 hydrocarbons. * Vannice's tabulati~n~~ of specific rates of methanation on Group VIII metals does not include 0 s . There is no reported value for the enthalpy of adsorption of CO on an 0 s surface, and therefore it is not possible to estimate the specific activity by interpolation of Vannice's volcano plot. Reduction in H2 at 473 K led to drastic spectral changes (fig. 7, spectrum 2).70 co HYDROGENATION CATALYSTS FROM OS, RU CLUSTERS CONCLUSIONS Supported 0 s and Ru catalysts with extremely high dispersion, prepared from O S ~ ( C O ) ~ ~ and RU~(CO)~~, respectively, and y-A1203, are active for CO hydrogenation, giving alkanes and alkenes.The Ru catalysts are about an order of magnitude more active than the 0 s catalysts and are an order of magnitude less selective for synthesis of C2 and higher hydrocarbons. The 0 s catalysts evidently consist of ensembles of three 0 s atoms bonded to near-neighbour oxygens of the A1203 support. We do not have enough evidence to establish a mechanism of the CO hydrogenation reaction on the present catalysts. It is interesting, however, to compare the results with those reported by Steinmetz and G e ~ f f r o y , ~ ~ who investigated the stepwise reduction of triosmium clusters in solution.The present results show that ensembles of varying sizes can be prepared on oxide surfaces. Ensemble size may affect catalytic activity and selectivity, but the data obtained with catalysts prepared from H20sCI, [which are more active and less selective for synthesis of C2 and higher hydrocarbons than those prepared from O S ~ ( C O ) ~ ~ ] suggest that the heterogeneity of the metal species and the composition of the support are also important. The work done in Munich was supported by the Deutsche Forschungsgemeinschaft, the Fonds der Chemischen Industrie and the Stiftung Volkswagenwerk, the work done in Delaware was supported by the National Science Foundation. A NATO grant provided support for the joint research. Y.Z. was a recipient of a grant from the German Academic Exchange Service.We express our gratitude to Dr. Storp, Bayer AG, Leverkusen, for his help with the X.p. spectroscopy. P. Biloen and W. M. H. Sachtler, Adv. Catal., in press. E. L. Muetterties and J. Stein, Chem. Rev., 1979, 79, 479. H. H. Nijs and P. A. Jacobs, J . Catal., 1980, 65, 328. R. L. Pruett, Science, 1981, 211, 11. B. C. Gates and J. Lieto, Chem. Tech., 1980, 195 and 248. M. Deeba, J. P. Scott, R. Barth and B. C. Gates, J. Catal., in press. M. Ichikawa, Bull. Chem. SOC. Jpn, 1978, 51, 2273. F. Hugues, B. Besson and J. M. Basset, J. Chem. SOC., Chem. Commun., 1980, 719. A. K. Smith, B. Besson, J. M. Basset, R. Psaro, A. Fusi and R. Ugo, J. Organomet. Chem., 1980,192, C31. lo B. Besson, B. Moraweck, A. K. Smith, J. M. Basset, R.Psaro, A. Fusi and R. Ugo, J. Chem. SOC., Chem. Commun., 1980, 569. l1 M. Deeba and B. C. Gates, J. Catal., 1981, 67, 303. l2 H. Knozinger and Y. Zhao, J, Catal., in press. l3 V, L. Kuznetsov, A. T. Bell and Y. I. Yermakov, J. Catal., 1980, 65, 374. l4 J. Robertson and G. Webb, Proc. R. SOC. London, Ser. A, 1974,341, 383. l5 C, R. Eady, B. F. G. Johnson and J. Lewis, J. Chem. Soc., Dalton Trans., 1975, 2606, l6 J. Lewis, personal communication, 1980. H. Knozinger, H. Stolz, H. Buhl, G. Clement and W. Meye, Chem. Ing. Tech., 1970, 42, 548; and H. Knozinger, Acta Cient. Venez., 1973, 24 Suppl. 2, 76. K. J. McQuade, J. P. Scott, R. Barth and B. C. Gates, in preparation. l9 D. A. Hucul and A. Brenner, J. Phys. Chem., 1981,85,496. 2o A. K. Smith, A. Theolier, J. M. Basset, R. Ugo, D. Commereuc and Y. Chauvin, J. Am. Chem. 21 E. L. Muetterties, T. N. Rhodin, E. Band, C. F. Brucker and W. R. Pretzer, Chem. Rev., 1979, 22 L. Brewer and G. M. Rosenblatt, Adv. High Temp. Sci., 1969, 2, 1. 23 J. H. Sinfelt, Rev. Mod. Phys., 1979, 51, 569. 24 E. B. Prestridge, G. H. Via and J. H. Sinfelt, J. Catal., 1977, 50, 115. 25 M. R. Churchill and B. G. de Boer, Inorg. Chem., 1977,16, 878. SOC., 1978, 100, 2590. 79, 91.H . KNOZINGER et al. 71 26 M. Primet, J. Chem. SOC., Faraday Trans. 1, 1978,74, 2570. 21 R. G. Woolley, Plat. Metals Rev., 1980, 24, 25. 28 R. G. Woolley, in Transition Metal Clusters, ed. B. F. G. Johnson (John Wiley & Sons, Chiches- L9 G. G. Low and A. T. Bell, J. Catal., 1979, 57, 397. 30 P. Biloen, J. N. Helle and W. M. H. Sachtler, J. Catal., 1979,58, 95. 31 R. D. Kelley and D. W. 6oodman, in Chemical Physics of Solid Surfaces and Heterogeneous Catalysts, ed. D. A. King and D. P. Woodruff (Elsevier, Amsterdam, to be published), vol. 4. 32 N. Sheppard and T. T. Nguyen, in Advances in Infrared Raman Spectroscopy, ed. R. J. H. Clark and R. E. Hester (Heyden, London, 1978), vol. 5, pp. 67-148. 33 G. H. Via, J. H. Sinfelt and F. W. Lytle, J. Chem. Phys., 1979, 71, 690. 34 Gy. Gati and H. Knozinger, 2. Phys. Chem. (Frankfurt am Main), 1972, 78, 243. 35 A. Zecchina, personal communication. 36 R. A. Dalla Betta and M. Shelef, J. Catal., 1977, 48, 111. 37 R. Ugo, R. Psaro, G. M. Zanderighi, J. M. Basset, A. Theolier and A. K. Smith, in Fundamen- tal Research in Homogeneous Catalysis, ed. M. Tsutsui (Plenum, New York, 1979), pp. 579-601. 38 M. A. Vannice, J. Catal., 1977, 50, 228. 39 G. R. Steinmetz and G. L. Geoffroy, J. Am. Chem. SOC., 1981, 103, 1278. ter, 1980), pp. 545-606.